Official repository for "Reasoning About the Unsaid: Misinformation Detection with Omission-Aware Graph Inference", AAAI 2026.

🕙 Preprint:

@misc{wang2025reasoningunsaidmisinformationdetection, title={Reasoning About the Unsaid: Misinformation Detection with Omission-Aware Graph Inference}, author={Zhengjia Wang and Danding Wang and Qiang Sheng and Jiaying Wu and Juan Cao}, year={2025}, eprint={2512.01728}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2512.01728}, } 

• 🛠️ Project: https://github.com/ICTMCG/OmiGraph
• 🔗 Paper: https://arxiv.org/abs/2512.01728
• 🏡 Home page: https://zhengjiawa.github.io/

"Learning from omission" for misinformation detection.

🌟 TL;DR:

This paper introduces OmiGraph, the first omission-aware misinformation detection framework. By recognizing that deception operates not only through what is explicitly stated but also through what is deliberately omitted, OmiGraph addresses a critical yet underexplored dimension of news deception.

Overview of OmiGraph.

We presented OmiGraph, the first omission-aware framework for misinformation detection. OmiGraph introduces omission-aware message-passing and aggregation that establishes holistic deception perception by integrating the omission contents and relations.

• constructs an omission-aware graph based on the contextual environment (a)
• omission-oriented relation modeling reasons over the graph nodes, identifying intra-source contextual dependencies and inter-source omission intents (b)
• an omission-guided message passing mechanism extracts omission-oriented deception features (c) to enhance conventional misinformation detectors

This research highlights how “Learning From Omission” offers a fundamentally novel and versatile paradigm. By demonstrating the feasibility and value of omission-aware modeling, OmiGraph opens new avenues for future research in trustworthy and interpretable misinformation mitigation solutions that can better serve the growing need in our increasingly complex media landscape.

📦OmiGraph ┣ 📂models ┃ ┣ 📜__init__.py ┃ ┣ 📜bert.py ┃ ┣ 📜layers.py ┃ ┗ 📜omi_graph.py ┣ 📂utils ┃ ┣ 📜dataset.py ┃ ┣ 📜misc.py ┃ ┗ 📜utils.py ┣ 📜README.md ┣ 📜engine.py ┣ 📜main.py ┗ 📜train.sh 

You can download the dataset from "Zoom Out and Observe: News Environment Perception for Fake News Detection (Sheng et al., ACL 2022)", and then place them to the folder ./data;

Run the shell script:

bash train.sh

Revise the storage locations for the model and results if needed.

If you find this repository useful, please cite our paper:

🕙 Preprint:

@misc{wang2025reasoningunsaidmisinformationdetection, title={Reasoning About the Unsaid: Misinformation Detection with Omission-Aware Graph Inference}, author={Zhengjia Wang and Danding Wang and Qiang Sheng and Jiaying Wu and Juan Cao}, year={2025}, eprint={2512.01728}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2512.01728}, }